"Prefer handling commands locally" doesn't work anymore

Configuration Voice Assistant
1

Hi everyone,

I’m using a Google Generative AI assistant with “Prefer handling commands locally” enabled, and since I upgraded to 2025.3 from 2025.2 (precisely to 2025.3.4 from 2025.2.2, I only upgrade once a month), all the commands are sent to Google and never handled locally. I tested with my esphome-based voice hardware, with the “Assist” text chat in the Android app, and with the option “Start debug run” “Text pipeline” in the web interface. All of them send all my command to the LLM. On the other hand, “Developer tools” “Assist” on the web interface does recognize the commands (either built-in or from custom_sentences). And finally, nothing is written in the logs about it, even when I enable debug logging on Google Generative AI integration options.

Anyone else got this problem? Any idea of how to debug it?

2

Try to use The versión 2.0 instead of 1.5 flash

3

I’m using 2.0 Flash Experimental

4

I have exactly the same problem. I’m on 2025.4.1.
When I switch to a non llm based pipeline it works.

5

I have this problem as well. Stopped working locally even though prefer handling commands locally is turned on. Is there a solution already, or should we create a bugreport?
Probably started with 2025.3.now on 2025.4.2

6

For me, it started working again when I upgraded to 2025.4.2 (as usual, I skipped all the updates since 2025.3.4, so I don’t know which update exactly fixed it)